Tapu Fini LoRA trained on NovelAI with Kohya training scripts using the DAdaptAdam optimizer at 512x512 resolution. Trained on a very low number of source images (29), so expect inconsistencies. Works quite well despite the low source image count. Has issues with vagina detail when prompting with (vagina), using (pussy) instead fixed the issue weirdly enough. That's likely a problem with my prompting setup and base model. Only does anthrofied Tapu Fini with the top half of her shell as a 'hat'. Should have wide compatibility with model mixes based on NovelAI, which is most of them. Trained on an all female dataset.
Description
Initial release
FAQ
Comments (12)
This one's for you, @WanoWano!
THIS IS AMAZING! 🤩
Thank you very much for making it, I hope it wasn't too annoying to work with, lol
It works very well too! You've just made my as well as the days of the other 3-4 Fini fans out there!
@WanoWano I actually didn’t have to spend too much time on this one. The secret to success was to crop out everything below the knees in my training images. That, combined with the few full body shots I had ended up yielding great results. It worked first try 🤯
@KomodoStar It sounds like it was simply meant to be, then 😏
And ooh, so that's how you avoided problems with the bottom half of the shell, that's smart.
Also, it's certainly much superior to the one I posted a little while ago 😂
@WanoWano I've been having to increasingly rely on cropping to get good training images as I've been working with fewer and fewer source images for the less popular mons. You can save a lot of images that wouldn't be suitable for LoRA training by cropping closer into the character and removing the irrelevant tags in post, which is what I did with the images that featured both Tapu Fini and Tapu Lele together. I felt like I was an angsty teen cutting my ex out of our old photos together, only thing missing is some angry marker scribbles over their face 😆
I saw the LoRA you posted, it was quite good for a first attempt! If you don't mind my advice, the main thing I'd suggest is to focus on having a consistent dataset. The more consistent the features in your training data, the more consistent the end result is. I saved my Rapidash model by snipping out 4-5 side profile images where the AI was interpreting the far ear on Rapidash's head as a second horn. Went from 2-4 horns on her head per gen down to the expected 1 lmao.
@KomodoStar The angsty photo cutting is hilarious 😂 Thank you for the advice on the dataset too 👍
it's good to see that you have supported this post very well,I hope to see you someday top 5 at least as a creator,I feel like a pokemon trainer catching them all,but in a more fun way,if someday my folder would weigh 100gb only of your Loras I wouldn't mind at all because they are very high quality Loras.
Of course I had to support it, this is the Lora I've wanted the most since I started using AI for generating images 🐟
@WanoWano it is great that you are supporting him,but what I would like is that others who download his Loras also contribute a little bit of Buzz to help the creator who deserves it,he makes very good quality loras and has made many for us,there are many people who download his loras and already,and do not contribute because they do not know or do not want to,I would like it to be more supported really because after all that little bit of Buzz helps(you can get a little free every day) I insist that we must support the good creators, to contribute a little is to be grateful and motivates the creator to continue creating,I repeat again(you can get 100 Buzz or more for free daily) I am sad to see many of your Loras with 0 Buzz without support,I for my part every time I get 100 Buzz I spend it on a Lora,I am supporting the evolutions of Eevee little by little meanwhile,I hope this comment gets to be read by other people.
@FireBlack85 It’s great that you’re enjoying the LoRAs I’m making and want to support me with buzz, but don’t go too far! Buzz is nice, but I don’t really have anything I want to spend it on 😅. I’m doing great financially, so I don’t need donations or anything like that. Just seeing reviews and comments is plenty enough for me ❤️
I might use the buzz I’ve pooled up in a tournament in the future, that could be fun 🤔
@KomodoStar it's not about economics here, it's about being grateful, how does a content creator feel when you see their videos or posts but you don't like, comment or anything? that's what I mean, every creator likes it when people praise their work, you may not have to spend the buzz now, but in the future I'm sure it will do some good.